4X The Unsurpassed Brian Reynolds' Alpha Centauri thread

[mastroego]

Yep. Maybe things like this could make it into a community driven patch, then we would make the AI build less worms.

What about an optional setting?
I could argue (actually, I do) that glaring conceptual flaws amount to bugs.

 

[Nim]

Any idea what causes the AI to build so many units that they run out of minerals ? Current game, sparta is going nuts, 41 units currently in 9 bases with 25 pop overall. And they are building new units in every single base.

Spoiler

 

[kyrub]

What about an optional setting?

Optional = 200% work on 200% of space you have to find in the code. Ask Yitzi (ac2 forums) about making things optional, he is a true master at that. All I can do is a reevaluation of the fonction, maybe there is a middle solution that will work even with minerals reduced. I have my doubts but we will see.

Any idea what causes the AI to build so many units that they run out of minerals ? Current game, sparta is going nuts, 41 units currently in 9 bases with 25 pop overall. And they are building new units in every single base.

Yes, I have noticed that as well. It seems a complex problem. One source is being tackled in the patch. Factions with focus on power and non-focus on wealth are susceptible to build too many units in their bases, especially if the bases have lower population. It has an asphyxiating effect. The victims are Spartans and Believers, though the latter ones tend to throw their units on the enemy and thus to avoid the problem. The patch should prevent and solve this. I have got an improved version for SMAX now on my HD.

The other source is a mystery. I have seen a Morgan base crowded suddenly with units. My secret tip is, there exists a re-homing fonction once a base is lost. The CP then (maybe) rehomes all unit mistakenly to 1 base only. Really not sure about that, but it is not the above case, since Morgan is cautious with the number of units.

 

[Nim]

Yup, Miri is also overdoing it. Morgan only has one base next to Santi so I can at least understand why he has 5 units defending it and is out of minerals. Though not why he hasn't built a rec tank.

 

[kyrub]

Some more questions: Has anybody seen AI build Centauri Preserve or Temple of Planet? The code seems strange.

No, really. It's like they do not know what armor is. Any AI should be taught to learn High Energy Chemistry before 2150.

And this!
What more Techs should be AI taught to learn? AIR combat and reactor techs are obvious and they are already in the code. Also note, that some techs are less useful for AI than for a human.

 

[Nevill]

Is the problem the numbers of defenders or the pure fact they persist with 1-1-1? Or both?

Both. They need more quality defenders. Maybe scout patrols shouldn't qualify as defenders with the passage of time?

Has anybody seen AI build Centauri Preserve or Temple of Planet?

I will have to check. I think I've seen Centauri Preserves before. Don't remember about ToP, though, by the time those are available the game is won long ago.

And this!

It could be remedied 'by hand' through assigning some crazy values to HEC, but the unfortunate by-product is that its cost in trading procedures skyrockets, with the AI willing to pay stupid amount of money or even trade bases for it. It is valuable, but not that valuable.

Centauri Ecology - obviously. Techs that grant Value civics - those are usually beneficial. Gene/EcoEng/EnvEcon line for lifting restrictions. Orbital Spaceflight (the AIs love rockets - and they are easy to use). Adv. Military Alrorithms (but that's on the way to Fusion reactor and Spaceflight).

If an AI often finds itself at war, it would not hurt to research Applied physics and Impact weaponry.

If they can be taught to use Clean reactor, then Bioengineering might be good to fix their SUPPORT woes.

Overall I don't see many critically important techs once the basics are taken care of.

 

[kyrub]

Both. They need more quality defenders. Maybe scout patrols shouldn't qualify as defenders with the passage of time?

Hey, not a bad idea, I take it as a solid base. The best solution, if AI would upgrade them, but that is a complex problem... and I have never seen units dismissed by AI, but there may be more findings later.

It could be remedied 'by hand' through assigning some crazy values to HEC, but the unfortunate by-product is that its cost in trading procedures skyrockets

Not necessary, AI is able to beeline for a tech. Only its "inner cost" goes up, when it chooses research. What a human player gets with "blind research", is the inner AI routine plus RNG.

Centauri Ecology - obviously. Techs that grant Value civics - those are usually beneficial. Gene/EcoEng/EnvEcon line for lifting restrictions. Orbital Spaceflight (the AIs love rockets - and they are easy to use). Adv. Military Alrorithms (but that's on the way to Fusion reactor and Spaceflight). Applied physics and Impact weaponry. Clean reactor

Other than value civics (good tip), they are all in the code. The trick may be in switching on only a few of the beelines at the right time. Shooting too many birds at the same time gives poor results. 2, max 3 beelines. Right now, it all works base on the AI focus, but it should be situational.

 

[Nevill]

The best solution, if AI would upgrade them, but that is a complex problem...

Do they upgrade units at all?

By the way, AI has Scouts as a separate Unit Plan assigned in the unit section (the one that defines formers, transports etc). Do you know what does that determine?

Shooting too many birds at the same time gives poor results. 2, max 3 beelines. Right now, it all works base on the AI focus, but it should be situational.

I have found out that giving AI different priorities (Conquer/Research/Build/Explore) makes them go for very different tech paths. Gaia, for example, love the EnvEcon line, while Morgan favors the Armor tech line for some reason, and University prefers Fusion and Pre-Sentient Algorithms and is often the first to get them.

The problem is that Gaia can go on for decades without bothering to learn Applied physics. I don't want to diminish factions' individuality, but there need to be at least some techs that add to a faction's basic survivability in the research queue. High Energy Chemistry is not that high price to pay, being only 2nd tier tech.

I am not sure I want each and every faction get it ASAP, though, but within first 9-10 techs? It would probably be worth it.

 

[kyrub]

Do they upgrade units at all?

The code is there, but I don't know why it does trigger so rarely. Maybe the lack of funds? AI spends horribly lot on hurrying everything, I have tried to make it selective. We will see. (Or try to give them lots of BC in editor and watch if they upgrade... Don't know.)

By the way, AI has Scouts as a separate Unit Plan assigned in the unit section (the one that defines formers, transports etc). Do you know what does that determine?

AI has a "recon" unit plan and uses it, esp. early. It looks fairly good with speeder chassis, but scout patrols are... well, scout patrols. I can tone down the production a bit. On the other hand, I really appreciate the game does not cheat (unlike many, many others) and the AI needs to scout, I'd rather not give this feature up. Recon units are used as auxiliary defense units, but only a given number can do so (lower threshold of ideal defense). I can also raise this lower threshold with time - AI will stop using them from turn X, for instance.

About them scout patrols: What about telling the AI to make Empathic Trance Scout patrols and obsolete normal ones as soon as it can? That might the AI a bit, both against the native life and worm-rushes. The moment I learn Secrets of the Human Brain, I obsolete my scout patrols.

Hmm, I guess obsolete comes with upgrading. Nice idea, we can surely desing a trance patrol or police patrol immediately. Whether AI can upgrade them, it must be seen. And tried.

I don't want to diminish factions' individuality, but there need to be at least some techs that add to a faction's basic survivability in the research queue. ...I am not sure I want each and every faction get it ASAP, though, but within first 9-10 techs? It would probably be worth it.

Actually, I think I can trigger certain things happening after X techs researched. I am cautious about AI individuality and flavour too, but we need more practical and situational choices from AI. The problem is, beelining after the war begins is in 90% cases too late.

 

[Nim]

On the other hand, I really appreciate the game does not cheat (unlike many, many others) and the AI needs to scout

What exactly do you mean by "needs to scout" ?
The AI certainly cheats, using full map knowledge to send colony pods to good base sites way outside its visibility. I've seen one of Zaks colony pods trying to make a 60+ turn trip when he had 2 bases while he didn't even have full visibility of his own base radius when I switched to him. The AI also loves to beeline for the jungle from halfway around the map, disregarding if it can even get there ( due to other bases or fungal towers f.ex. ).
And from the way they catch me on the seas I'd bet a lot on them knowing exactly where all units are, they just can't exploit this on land due to rocky and fungus tiles and slower movement in general.

 

[kyrub]

Hm. You are probably right (seems I was naive ).
I have checked and the "recon" unit plan is probably reserved to one and only unit, Scout patrol. The unit cannot be upgraded, because it does not share plan with others. We will have to change the internal unit plan after X turns. That's all, then it should work.

Other than that, the upgrade routine is bugged and suboptimal. Have anybody seen Spartans dip into negative credits? It seems it can happen on highest difficulties. Anyway, it's easy to repair.

 

[Nevill]

I've seen one of Zaks colony pods trying to make a 60+ turn trip when he had 2 bases while he didn't even have full visibility of his own base radius when I switched to him.

There is probably a bug in pathfinding. Sometimes they send units to the other half of the map, seemingly without purpose. It is highly annoying to see them do this with their second or third Colony Pod.

But yes, the AI definitely cheats. Stealth units (or units in fungus) do not work against it, for example, as it knows where any unit is on the map.

Have anybody seen Spartans dip into negative credits?

The most common occurence of AI dipping into the negatives could be observed when you sell them techs, especially when pacted. There is no check to see if they have enough.

Confirming that AI goes into negatives is nearly impossible, as once you hit negative credits you facilities will overload. The effect is the same as when you scrap them yourself for energy. Essentially, any negative credits should fix themselves on the next turn... at a cost to AI's development. The above case with tech trading can only be verified because it is initiated during the player's turn.

The game is a mess.

 

[Nim]

I doubt that colony pods going all over the map is a bug in pathfinding, the AI simply doesn't evaluate base sites like we do. They seem to totally ignore distance and the benefits you get when putting bases closer together. And sometimes they go full retard, like Otkrietia-Discovery here that is 10+ squares away from the next uni base and working a 1 nut/energy square

Spoiler

 

[Nevill]

I wonder why do jungles have such density of bases. Is extra food all there is to it? Or does AI calculate the production value of all nearby squares, see it going through the roof from all the +1 food tiles, and decides that this is the best spot ever?

At least Yang does something right, so maybe there are relevant bits in his AI code.

Another big offender (and I think I wrote about this one) is that sometimes AI colony pods forget what they were doing. The AI sends those to a location, seemingly to start a base, and once they arrive, it sends them elsewhere, or possibly even back where they came from.

 

[Nim]

Another big offender (and I think I wrote about this one) is that sometimes AI colony pods forget what they were doing. The AI sends those to a location, seemingly to start a base, and once they arrive, it sends them elsewhere, or possibly even back where they came from.

Could it be that the AI never reevaluates movement orders ? Only once they either arrive or get stuck it remembers that unit ?

 

[Nevill]

That too, yes. But if they bothered to move all the way with the obvious intention to set up a base, why move away upon arrival?

Especially if they do found a base on that very same spot some 15 years later.

 

[Nim]

Enemy/native units showing up would be the only thing that seems valid there, otherwise

The game is a mess.

 

[Nevill]

Enemy/native units showing up would be the only thing that seems valid there

Nope. No units. Just some messed up pathfinding.

Another example would be an island AI, trying to get its troops to another continent. On foot.

indeed.

 

[kyrub]

I wonder why do jungles have such density of bases. Is extra food all there is to it? Or does AI calculate the production value of all nearby squares, see it going through the roof from all the +1 food tiles, and decides that this is the best spot ever?

Well, it's a common knowledge that AI does best on the Jungle terrain. Any AI simply loves this terrain. In fact, in my patch, they love it even more than in vanilla. And I think it's a valid strategy, since the AI who takes the jungle tends to win the game.
All AIs have a deep appreciation of landmarks. I may have to tone it down, if necessary. And yes, AI takes distance into account, to some extent.

Nim, are these pictures from SMAC 444p patch or from an unpatched game?
What happened on the map, is a bit hard to tell. Either there was another base between them, that went down to a worm (my tip). Even if not so, are there many good places for a base? AI takes arid squares for 0 value and it actively dislikes fungus in neighbourhood, and rightly so, since it is unable to protect its own formers against worms, with notoriety.

Nevill said: ↑Another big offender (and I think I wrote about this one) is that sometimes AI colony pods forget what they were doing. The AI sends those to a location, seemingly to start a base, and once they arrive, it sends them elsewhere, or possibly even back where they came from.

Could it be that the AI never reevaluates movement orders ? Only once they either arrive or get stuck it remembers that unit ?

I know about it. It is a mystery. I once watched a colony pod circling on 3 squares, one step there, two steps back. 40 turns and on and on. I was trying to solve it. No success, so far. My theory is, it has something to do with the equation and the distance, and, probably, rounding down after dividing. So the equation may have two results, when on a different square.

 

[Nim]

Nim, are these pictures from SMAC 444p patch or from an unpatched game?

SMAX, Yitzi patch, 3.4 I think.

What happened on the map, is a bit hard to tell. Either there was another base between them, that went down to a worm (my tip). Even if not so, are there many good places for a base? AI takes arid squares for 0 value and it actively dislikes fungus in neighbourhood, and rightly so, since it is unable to protect its own formers against worms, with notoriety.

You should be able to click on the pic to enlarge it, it's actually 1920x1200.
Nah, there was no base inbetween.
The problem is not only the distance to Zaks other bases, but also that there is >1< square in base radius that has minerals without t-forming, the mineral resources one. Which had no nutrients so the AI assigned its worker to a 1 nut/1 energy square.
Now for better spots, to the left of freshwater sea that looks like a 2/1/0 square that might get a base going. To the top right of monitoring station would be another good spot. If the AI knew how to terraform there would be quite a few more.

 

[kyrub]

SMAX, Yitzi patch, 3.4 I think.

I am relieved to hear it. I believe the AI patch does slightly better job at choosing the right place, but you will see. Part of the fungu-phobia was removed, the low distance is a bit more important. And it does better job at making resources worthwile with terraforming.

 

[Nim]

While we're on stuff being messed up, anyone can explain this ?

Spoiler

Miri moves the scout patrols out of all her bases ( Holy Fire doesn't have one yet) and they all have movement orders to squares that are each 2 tiles away from each base. WTF.

Bonuspic, check this colony pod

Spoiler

 

[Nevill]

All AIs have a deep appreciation of landmarks. I may have to tone it down, if necessary.

Nah. It does a good job of colonizing landmarks. If anything, I'd like it to appreciate regular spots more. If it does not terraform it, how else would the land improve?

 

[kyrub]

One of Yitzi's patch most welcome features was the option to mod which units produce pacifist drones and under what conditions. There were requests to have a separate mode specifically for the AI. If implemented, AI opponents could be set to not get a penalty for planes - or if one wants to, for any units at all.

I have looked at Punishment sphere code today and one thing springs to my mind.
What if the AI uses the PS + rehoming units? I dislike the tactics and never use it, personnally, since it feels like an exploit, too easy to do. But if we are at a stage to consider extra favourable rules for the AI, it's probably better to give it a regular, albeit cheesy tactics. And then it's player's choice, if he uses the rehoming as well.

The thing is, it may be actually relatively easy to implement. Yitzi even cleared some code in the area, so we could just put there a small bypass, search for a base with PS and rehome the units when there is enough minerals. Still some work, but definitely doable. And what's better, AI can rehome the unit in the moment it enters enemy territory. So, no drones will appear, no SE reconsideratoin. The more I think about it, the more it sounds as reallly possible.

 

[Nevill]

What if the AI uses the PS + rehoming units? I dislike the tactics and never use it, personnally, since it feels like an exploit, too easy to do.

Well, you lose a base for any kind of meaningful development (you do not want to develop a crippled one), and you can't really rehome too much units to it anyway, unless you are using crawlers. Which are broken by themselves, or so I believe and try to argue for in this very same thread.

It would be a solution, yes. Actually, it would be best if AI could be taught to set some bases to serve as auxillary ones to alleviate its SUPPORT issues. They wouldn't produce anything, just simple bases on two forests or a single borehole, to house 5-7 units. I do this fairly often, to reduce the strain on my 'better' bases.

I did not know you could actually code that, though. I assumed you could only fix/improve what is already in place.

I suppose it could help with air units and ships. but I am afraid nothing could be done for land units. You don't know which one of them will create pacifist drones, as they don't do that while inside your own borders. To be able to remedy that, you would have to teach AIs to attack only with units homed to bases with Punishment Spheres, and I am fairly certain it is beyond the realm of possibility to code this in without breaking anything.